R-CCS Cafe 第201回 第2部

Towards the end of Moore's law, we need to develop not only new hardware, but also new algorithms and applications. In this study, we propose an innovative method for computational science for sustainable promotion of scientific discovery by supercomputers in the Exascale Era by combining (Simulation + Data + Learning (S+D+L)), where ideas of data science and machine learning are introduced to computational science. The BDEC system (Big Data & Extreme Computing), which is scheduled to be introduced to the Tokyo University in 2021, is a Hierarchical, Hybrid, Heterogeneous (h3) system, which consists of computing nodes for computational science and those for data science/machine learning. In this study, we consider the BDEC as the platform for integration of (S+D+L), develop an innovative software platform “h3-Open-BDEC” for integration of (S+D+L), and evaluate the effects of integration of (S+D+L) on the BDEC. The h3-Open-BDEC is designed for extracting the maximum performance of the supercomputers with minimum energy consumption focusing on (1) innovative method for numerical analysis with high-performance/high-reliability/power-saving based on the new principle of computing by adaptive precision, accuracy verification and automatic tuning, and (2) Hierarchical Data Driven Approach (hDDA) based on machine learning. Integration of (S+D+L) by h3-Open-BDEC enables significant reduction of computations and power consumption, compared to those by conventional simulations. In this talk, recent progress in accuracy verification for sparse linear solvers will be discussed.